Skip to content
mlmentorship
mlmentorship field guide2026 edition

ML from Primitives · senior to principal

The ML Interview Field Guide

A free visual field guide for senior ML, AI systems, and frontier-lab interviews.

Begin visual review Build my prep plan

391 visual lessons · 10 ordered books · free to read

One idea, seen

A matrix is a map, not a table of numbers.

Map each basis vector to a column. Keep the input's weights. The geometry explains the rank, range, and the direction that disappears.

See the full visual lesson
A matrix maps basis vectors to its columns An input vector x equals two e1 minus e2 maps across a matrix to two times column a1 minus column a2. The output columns lie on one line, so the nonzero input direction collapses to zero. INPUT SPACE 2e1−e2x = 2e1 − e2 same weights OUTPUT SPACE a1 = (1, 1)a2 = (2, 2) = 2a1 Ax = 2a1 − a2 = 0columns share one line A matrix maps basis vectors to its columns The input vector x equals two e1 minus e2. Applying the same weights to the matrix columns gives Ax equals two a1 minus a2 equals zero because both output columns share one line. 1 · INPUT SPACE x = 2e1 − e2 2e1−e2 same weights 2 · OUTPUT SPACE a1 = (1, 1)a2 = 2a1 Ax = 2a1 − a2 = 0one output line
Read it this way: the matrix acts on the basis first. Once the columns are visible, the “rank-one” result is a shape, not a memorized phrase.
Start here

Choose the shortest route from what you already know.

You do not need to read the library front to back.

The curriculum

10 books, ordered from foundations to interview execution.

Open a shelf, then choose a book or start its first visual.

Core MLBuild the technical base used across applied, research, and engineering interviews.
  1. I

    ML foundations

    Math, probability, classical machine learning, deep learning, and the core questions that test them.

    9 chapters · 62 entries
    Start visually
  2. II

    Model training and research

    Optimization, reliable experiments, implementation, debugging, and research judgment.

    9 chapters · 52 entries
    Start visually
  3. III

    Evaluation and product ML

    Metrics, experimental validity, calibration, product decisions, and production evaluation.

    4 chapters · 25 entries
    Start visually
Frontier AI systemsStudy language models, post-training, agents, accelerators, and distributed systems.
  1. IV

    LLMs, agents, and post-training

    Transformer internals, inference, retrieval, evaluation, agents, alignment, and post-training.

    7 chapters · 42 entries
    Start visually
  2. V

    ML systems and infrastructure

    Accelerators, distributed training, inference systems, reliability, cost, and full ML architecture.

    6 chapters · 31 entries
    Start visually
Specialist tracksAdd only the specialist subject required by the role and team.
  1. VI

    Retrieval, ranking, and recommendations

    Embeddings, candidate generation, ranking, search metrics, cold start, and feedback loops.

    4 chapters · 23 entries
    Start visually
  2. VII

    Reinforcement learning and robotics

    Sequential decisions, value and policy methods, environments, rewards, and robotics policy learning.

    3 chapters · 13 entries
    Start visually
  3. VIII

    Vision, language, and speech

    Visual models, multimodal systems, sequence modeling, natural language, and speech.

    4 chapters · 21 entries
    Start visually
Interview executionPrepare role choice, project evidence, behavioral judgment, and senior-level communication.
  1. IX

    Interview and career practice

    Role choice, level calibration, project stories, behavioral judgment, and long-form field guides.

    3 chapters · 15 entries
    Start visually
  2. X

    Coding interview practice

    A visual-first coding field guide for data structures, algorithms, and practical AI coding. Learn each problem by seeing the state it preserves, the move it makes, and the invariant that makes the move safe.

    11 chapters · 107 entries
    Start visually
Built from practice

Specific, current, and honest about scope.

Written by Hamidreza Saghir, Principal Applied Scientist at Microsoft, with earlier ML engineering, applied-science, and research roles at X, Amazon, and Borealis AI. The site uses public process evidence, never leaked prompts or job-outcome promises. About the author and project.